Update app.py

app.py

@@ -1,159 +1,98 @@
-import os
-os.environ["NUMBA_DISABLE_CACHE"] = "1"
-
 import gradio as gr
-from docx import Document
 from TTS.api import TTS
 import tempfile
+import docx
+
+# Voice models dictionary with metadata on whether they support multi-speaker
+VOICE_MODELS = {
+    "LJSpeech (Standard Female)": {
+        "model_name": "tts_models/en/ljspeech/vits",
+        "multi_speaker": False
+    },
+    "VCTK (Multi-speaker English)": {
+        "model_name": "tts_models/en/vctk/vits",
+        "multi_speaker": True
+    }
+}
 
-# Embedding the metadata directly into the script
+# Embedded short speaker metadata (from your CSV)
 SPEAKER_METADATA = {
-    300: {"audio_id": 1, "age": 23, "gender": "F", "accent": "American"},
-    271: {"audio_id": 1, "age": 19, "gender": "M", "accent": "Scottish"},
-    287: {"audio_id": 1, "age": 23, "gender": "M", "accent": "English"},
-    262: {"audio_id": 1, "age": 23, "gender": "F", "accent": "Scottish"},
-    284: {"audio_id": 1, "age": 20, "gender": "M", "accent": "Scottish"},
-    297: {"audio_id": 1, "age": 20, "gender": "F", "accent": "American"},
-    227: {"audio_id": 1, "age": 38, "gender": "M", "accent": "English"},
-    246: {"audio_id": 1, "age": 22, "gender": "M", "accent": "Scottish"},
-    225: {"audio_id": 1, "age": 23, "gender": "F", "accent": "English"},
-    259: {"audio_id": 1, "age": 23, "gender": "M", "accent": "English"},
-    252: {"audio_id": 1, "age": 22, "gender": "M", "accent": "Scottish"},
-    231: {"audio_id": 1, "age": 23, "gender": "F", "accent": "English"},
-    266: {"audio_id": 1, "age": 22, "gender": "F", "accent": "Irish"},
-    241: {"audio_id": 1, "age": 21, "gender": "M", "accent": "Scottish"},
-    312: {"audio_id": 1, "age": 19, "gender": "F", "accent": "Canadian"},
-    329: {"audio_id": 1, "age": 23, "gender": "F", "accent": "American"},
-    232: {"audio_id": 1, "age": 23, "gender": "M", "accent": "English"},
-    305: {"audio_id": 1, "age": 19, "gender": "F", "accent": "American"},
-    311: {"audio_id": 1, "age": 21, "gender": "M", "accent": "American"},
-    301: {"audio_id": 1, "age": 23, "gender": "F", "accent": "American"},
-    304: {"audio_id": 1, "age": 22, "gender": "M", "accent": "NorthernIrish"},
-    310: {"audio_id": 1, "age": 21, "gender": "F", "accent": "American"},
-    260: {"audio_id": 1, "age": 21, "gender": "M", "accent": "Scottish"},
-    315: {"audio_id": 1, "age": 18, "gender": "M", "accent": "American"},
-    374: {"audio_id": 1, "age": 28, "gender": "M", "accent": "Australian"},
-    364: {"audio_id": 1, "age": 23, "gender": "M", "accent": "Irish"},
-    269: {"audio_id": 1, "age": 20, "gender": "F", "accent": "English"},
-    345: {"audio_id": 1, "age": 22, "gender": "M", "accent": "American"},
-    326: {"audio_id": 1, "age": 26, "gender": "M", "accent": "Australian"},
-    343: {"audio_id": 1, "age": 27, "gender": "F", "accent": "Canadian"},
-    230: {"audio_id": 1, "age": 22, "gender": "F", "accent": "English"},
-    376: {"audio_id": 1, "age": 22, "gender": "M", "accent": "Indian"},
-    240: {"audio_id": 1, "age": 21, "gender": "F", "accent": "English"},
-    298: {"audio_id": 1, "age": 19, "gender": "M", "accent": "Irish"},
-    272: {"audio_id": 1, "age": 23, "gender": "M", "accent": "Scottish"},
-    248: {"audio_id": 1, "age": 23, "gender": "F", "accent": "Indian"},
-    264: {"audio_id": 1, "age": 23, "gender": "F", "accent": "Scottish"},
-    250: {"audio_id": 1, "age": 22, "gender": "F", "accent": "English"},
-    292: {"audio_id": 1, "age": 23, "gender": "M", "accent": "NorthernIrish"},
-    237: {"audio_id": 1, "age": 22, "gender": "M", "accent": "Scottish"},
-    363: {"audio_id": 1, "age": 22, "gender": "M", "accent": "Canadian"},
-    313: {"audio_id": 1, "age": 24, "gender": "F", "accent": "Irish"},
-    285: {"audio_id": 1, "age": 21, "gender": "M", "accent": "Scottish"},
-    268: {"audio_id": 1, "age": 23, "gender": "F", "accent": "English"},
-    302: {"audio_id": 1, "age": 20, "gender": "M", "accent": "Canadian"},
-    261: {"audio_id": 1, "age": 26, "gender": "F", "accent": "NorthernIrish"},
-    336: {"audio_id": 1, "age": 18, "gender": "F", "accent": "SouthAfrican"},
-    288: {"audio_id": 1, "age": 22, "gender": "F", "accent": "Irish"},
-    226: {"audio_id": 1, "age": 22, "gender": "M", "accent": "English"},
-    277: {"audio_id": 1, "age": 23, "gender": "F", "accent": "English"},
-    360: {"audio_id": 1, "age": 19, "gender": "M", "accent": "American"},
-    257: {"audio_id": 1, "age": 24, "gender": "F", "accent": "English"},
-    254: {"audio_id": 1, "age": 21, "gender": "M", "accent": "English"},
-    339: {"audio_id": 1, "age": 21, "gender": "F", "accent": "American"},
-    323: {"audio_id": 1, "age": 19, "gender": "F", "accent": "SouthAfrican"},
-    255: {"audio_id": 1, "age": 19, "gender": "M", "accent": "Scottish"},
-    249: {"audio_id": 1, "age": 22, "gender": "F", "accent": "Scottish"},
-    293: {"audio_id": 1, "age": 22, "gender": "F", "accent": "NorthernIrish"},
-    244: {"audio_id": 1, "age": 22, "gender": "F", "accent": "English"},
-    245: {"audio_id": 1, "age": 25, "gender": "M", "accent": "Irish"},
-    361: {"audio_id": 1, "age": 19, "gender": "F", "accent": "American"},
-    314: {"audio_id": 1, "age": 26, "gender": "F", "accent": "SouthAfrican"},
-    308: {"audio_id": 1, "age": 18, "gender": "F", "accent": "American"},
-    229: {"audio_id": 1, "age": 23, "gender": "F", "accent": "English"},
-    341: {"audio_id": 1, "age": 26, "gender": "F", "accent": "American"},
-    275: {"audio_id": 1, "age": 23, "gender": "M", "accent": "Scottish"},
-    263: {"audio_id": 1, "age": 22, "gender": "M", "accent": "Scottish"},
-    253: {"audio_id": 1, "age": 22, "gender": "F", "accent": "Welsh"},
-    299: {"audio_id": 1, "age": 25, "gender": "F", "accent": "American"},
-    316: {"audio_id": 1, "age": 20, "gender": "M", "accent": "Canadian"},
-    282: {"audio_id": 1, "age": 23, "gender": "F", "accent": "English"},
-    362: {"audio_id": 1, "age": 29, "gender": "F", "accent": "American"},
-    294: {"audio_id": 1, "age": 33, "gender": "F", "accent": "American"},
-    274: {"audio_id": 1, "age": 22, "gender": "M", "accent": "English"},
-    279: {"audio_id": 1, "age": 23, "gender": "M", "accent": "English"},
-    281: {"audio_id": 1, "age": 29, "gender": "M", "accent": "Scottish"},
-    286: {"audio_id": 1, "age": 23, "gender": "M", "accent": "English"},
-    258: {"audio_id": 1, "age": 22, "gender": "M", "accent": "English"},
-    247: {"audio_id": 1, "age": 22, "gender": "M", "accent": "Scottish"},
-    351: {"audio_id": 1, "age": 21, "gender": "F", "accent": "NorthernIrish"},
-    283: {"audio_id": 1, "age": 24, "gender": "F", "accent": "Irish"},
-    334: {"audio_id": 1, "age": 18, "gender": "M", "accent": "American"},
-    333: {"audio_id": 1, "age": 19, "gender": "F", "accent": "American"},
-    295: {"audio_id": 1, "age": 23, "gender": "F", "accent": "Irish"},
-    330: {"audio_id": 1, "age": 26, "gender": "F", "accent": "American"},
-    335: {"audio_id": 1, "age": 25, "gender": "F", "accent": "NewZealand"},
-    228: {"audio_id": 1, "age": 22, "gender": "F", "accent": "English"},
-    267: {"audio_id": 1, "age": 23, "gender": "F", "accent": "English"},
-    273: {"audio_id": 1, "age": 18, "gender": "F", "accent": "English"}
+    "225": {"age": 23, "gender": "F", "accent": "English"},
+    "226": {"age": 22, "gender": "M", "accent": "English"},
+    "227": {"age": 38, "gender": "M", "accent": "English"},
+    "228": {"age": 22, "gender": "F", "accent": "English"},
+    "229": {"age": 23, "gender": "F", "accent": "English"},
+    "230": {"age": 22, "gender": "F", "accent": "English"},
+    "231": {"age": 23, "gender": "F", "accent": "English"},
+    "232": {"age": 23, "gender": "M", "accent": "English"},
+    "233": {"age": 23, "gender": "F", "accent": "English"},
+    "234": {"age": 22, "gender": "F", "accent": "Scottish"}
+    # Add more as needed
 }
 
-# Load the TTS model
-tts = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", gpu=False)
-
-# Extract speakers from metadata
-def extract_speakers(voice_selection):
-    speaker_choices = []
-    for speaker_id, metadata in SPEAKER_METADATA.items():
-        if voice_selection == "english":
-            speaker_choices.append((
-                str(speaker_id),
-                f"p{speaker_id} ({metadata['gender']}, {metadata['accent']}, {metadata['age']} yrs)"
-            ))
-    return speaker_choices
-
-# Update the speaker dropdown based on selected voice
-def update_speaker_dropdown(voice_selection):
-    speaker_choices = extract_speakers(voice_selection)
-    visible = bool(speaker_choices)
-    default = speaker_choices[0][0] if speaker_choices else None
-    return gr.Dropdown.update(
-        choices=speaker_choices,
-        visible=visible,
-        value=default
-    )
+# Pre-format speaker dropdown choices
+SPEAKER_CHOICES = [
+    (sid, f"p{sid} ({data['gender']}, {data['accent']}, {data['age']} yrs)")
+    for sid, data in SPEAKER_METADATA.items()
+]
+
+# Model cache
+MODEL_CACHE = {}
+
+def load_tts_model(model_key):
+    if model_key in MODEL_CACHE:
+        return MODEL_CACHE[model_key]
+    model_info = VOICE_MODELS[model_key]
+    tts = TTS(model_name=model_info["model_name"], gpu=False)
+    MODEL_CACHE[model_key] = tts
+    return tts
+
+def extract_text_from_docx(file):
+    doc = docx.Document(file)
+    return "\n".join([para.text for para in doc.paragraphs if para.text.strip()])
+
+def generate_audio(voice_key, speaker_id, docx_file):
+    text = extract_text_from_docx(docx_file)
+    tts = load_tts_model(voice_key)
+    kwargs = {}
 
-# Generate speech to text
-def generate_audio(voice_selection, speaker_selection, text_input):
-    speaker_id = int(speaker_selection)
-    temp_file = tempfile.mktemp(suffix=".wav")
-    tts.tts_to_file(text_input, temp_file, speaker=speaker_id)
-    return temp_file
+    if VOICE_MODELS[voice_key]["multi_speaker"]:
+        kwargs["speaker"] = speaker_id
+
+    output_path = tempfile.mktemp(suffix=".wav")
+    tts.tts_to_file(text=text, file_path=output_path, **kwargs)
+    return output_path
+
+def update_speaker_visibility(voice_key):
+    visible = VOICE_MODELS[voice_key]["multi_speaker"]
+    return gr.update(visible=visible)
 
-# Gradio interface
 with gr.Blocks() as demo:
+    gr.Markdown("## DOCX to Speech with Speaker Selection")
+
     with gr.Row():
-        with gr.Column():
-            voice_dropdown = gr.Dropdown(
-                choices=["english", "other"], 
-                label="Select Voice",
-                value="english"
-            )
-            speaker_dropdown = gr.Dropdown(
-                label="Select Speaker",
-                visible=False
-            )
-            text_input = gr.Textbox(label="Enter text")
-            audio_output = gr.Audio(label="Generated Audio")
-            
-    voice_dropdown.change(
-        fn=update_speaker_dropdown,
-        inputs=voice_dropdown,
-        outputs=speaker_dropdown
+        voice_dropdown = gr.Dropdown(
+            choices=list(VOICE_MODELS.keys()),
+            value="LJSpeech (Standard Female)",
+            label="Select Voice"
+        )
+
+        speaker_dropdown = gr.Dropdown(
+            choices=SPEAKER_CHOICES,
+            label="Select Speaker",
+            visible=False
+        )
+
+    docx_input = gr.File(label="Upload .docx File", file_types=[".docx"])
+    generate_btn = gr.Button("Generate Audio")
+    audio_output = gr.Audio(label="Output Audio")
+
+    voice_dropdown.change(fn=update_speaker_visibility, inputs=voice_dropdown, outputs=speaker_dropdown)
+
+    generate_btn.click(
+        fn=generate_audio,
+        inputs=[voice_dropdown, speaker_dropdown, docx_input],
+        outputs=audio_output
     )
-    
-    generate_button = gr.Button("Generate Audio")
-    generate_button.click(generate_audio, inputs=[voice_dropdown, speaker_dropdown, text_input], outputs=audio_output)
 
 demo.launch()